Property taxation, bounded rationality and housing prices

Department of Economics

Working Paper 2014:4

Property taxation, bounded rationality and housing prices

Mikael Elinder and Lovisa Persson

Department of Economics Working paper 2014:4

Uppsala University May 2014

P.O. Box 513 ISSN 1653-6975

SE-751 20 Uppsala Sweden

Fax: +46 18 471 14 78

Property taxation, bounded rationality and housing prices

Mikael Elinder and Lovisa Persson

Papers in the Working Paper Series are published on internet in PDF formats.

Download from http://www.nek.uu.se or from S-WoPEC http://swopec.hhs.se/uunewp/

Property taxation, bounded rationality and housing prices ^∗

Mikael Elinder

Lovisa Persson

May 22, 2014

Abstract

In 2008, the Swedish property tax was reformed and a cap on yearly tax liabilities was introduced. A large fraction of owner occupied houses was subject to a substantial decrease in the tax. When the reform was announced, most analysts projected - in line with tax capitalization theory - that the tax decrease would lead to significant increases in house prices. We estimate price responses and capitalization degrees, using various DID strategies, in which the price dynamics of houses that were subject to a generous tax reduction are compared to the price dynamics of houses with a more modest reduction. Our results are largely inconsistent with capitalization theory. For the majority of properties, we find no evidence that the tax cut led to increases in house prices. However, we find evidence of partial capitalization in sub-markets with highly valued properties, highly educated citizens and were it is especially difficult to increase supply. We argue that theories of bounded rationality can help explain why house buyers may fail to take a tax decrease into account in the valuation of houses.

Keywords: announcement effects, capitalization, financial literacy, housing market, inattention, saliency

JEL codes: D01, D03, D04, D12, H22, H24, R21, R38

∗We would like to thank Aron Berg and Nina ¨Ohrn Karlsson for excellent research assistance, Niclas Berggren, Per Engstr¨om, Henrik Jordahl, Che-Yuan Liang, Eva M¨ork, Erik Spector, Oskar Tysklind and seminar participants at The Research Institute of Industrial Economics (IFN), The Ministry of Finance and Uppsala University for valuable comments, and the Swedish Research Council for financial support.

†Department of Economics and Uppsala Center for Fiscal Studies (UCFS), Uppsala University, and the Research Institute of Industrial Economics (IFN); Email: mikael.elinder@nek.uu.se

‡Department of Economics and Uppsala Center for Fiscal Studies (UCFS), Uppsala University; Email:

lovisa.persson@nek.uu.se

1 Introduction

In 2008, property taxes were substantially reduced for a large share of Swedish owner occupied residential properties (henceforth referred to as just houses or properties). In line with standard capitalization models, most analysts predicted that the tax cut would lead to increases in house prices. In this paper we estimate to what extent the reduction in the Swedish property tax was capitalized into house prices. To identify causal effects of the tax cut, we compare the price dynamics of houses that received a small decrease in taxes due to the reform, with the price dynamics of houses that received a larger decrease.

Since both groups had very similar price developments prior to the reform, we find our strategy reliable. Overall, our results suggest that for the vast majority of properties the reduction in the property tax had no effect on house prices, neither at the time of announcement, nor in the time period after implementation. We discuss whether the absence of general price responses is due to bounded rationality of house buyers. We find substantial capitalization only in sub-markets characterized by highly valued houses, large reductions in the property tax and highly educated citizens. Consequently, we find it plausible that the property tax is neglected in the valuation of houses by a large portion house buyers – as a way of simplifying a complex decision.

In capitalization models (see for instance Oates (1969) and Yinger (1982)), a prospec- tive house buyer considers both house characteristics and factors affecting the cost of living when deciding what price to offer the seller. The market value of the current and future stream of these characteristics determine the market price. When a property tax is lowered, buyers take into account that the cost of living has decreased and they will therefore be willing to make higher price offers. If the supply of land and housing is fixed, the market price will increase with the net present value of the reduction in present and future tax payments (Oates (1969)).

If the housing market is efficient, and individuals use all relevant information in order to value a house, prices will change immediately when new information about future tax changes are made public (Ross and Yinger (1999)).

Capitalization models rely on strong assumptions about the degree of sophistication among house buyers as they calculate the net present value of all aspects of a property when determining the value. This includes aspects like location, size, attractiveness, main- tenance costs, taxes and much more. However, for many individuals, buying a house is a new type of decision problem, of which they have limited experience. There are two poten- tial sources of optimization errors in this situation. First, recent studies have documented that large groups of consumers lack the financial ability to make rational investment deci- sions (see Lusardi and Mitchell (2014) for a review of the financial literacy literature). It

1If the property tax is viewed as a tax on capital, full capitalization is expected when the property tax is uniform within a nation and capital cannot move freely between sectors, see Mieszkowski (1972) for the

“capital tax” approach.

may be difficult for many individuals to manage to perform the task that capitalization theory demands from them, namely to assess the net present value of lower property taxes.

Second, optimization errors may also be catalyzed by the characteristics of the decision making environment. When there are many aspects to take into account (competing stim- uli), such as when buying a house, potentially important information could be neglected by the consumer in order to simplify the decision, see DellaVigna (2009). A boundedly rational consumer may neglect less salient or less important (such as small costs) aspects, and instead focus on the prioritized characteristics or the heaviest costs. The property tax could then be a relatively non-salient aspect for most houses buyers, especially if the prop- erty tax liability is low.

When taxes are not salient for the consumer, conclusions about tax incidence need to be modified, and Chetty (2009) therefore derives formulas that char- acterize the welfare implications in such a situation. In an experiment where commodity prices in a grocery store are varyingly displayed with or without taxes included, Chetty et al. (2009) show that non-saliency in commodity taxes increases demand by 8 percent.

In this paper we argue that both limited financial literacy and the relative non-saliency of the property tax are likely to mitigate house price responses to the tax change.

The empirical literature on property tax capitalization has typically found full or par- tial capitalization in house prices (see reviews by Ross and Yinger (1999), Sirmans et al.

(2008) and Hilber (2011)). Most studies have estimated property tax capitalization in a local government context with cross-sectional variation in property tax rates. Two impor- tant identification problems arise when studying the property tax in a local public sector context. First, the level of public goods is positively correlated with the property tax level, and both the tax level and the public goods level independently affect the house price level. If the tax differential that remain after controlling for public services is equal to the house price differential, there is said to be full capitalization of the property tax.

There are however severe difficulties concerning how to measure the level of public services in a way as to avoid biased estimates. Second, when local governments set their own tax rate, areas with a high house price level, all else equal, are able to set a lower tax rate to collect a certain amount of tax revenues. This creates a simultaneity bias between the property tax rate and house prices. Both of these identification issues have been known and discussed since the seminal paper by Oates (1969).

The above identification issues are dealt with in Cushing (1984) by comparing blocks of housing on opposite sides of the boarder of two jurisdictions with different property tax rates. This reduces the endogeneity problem since the compared houses largely share the same type of services and neighbourhood amenities. The results indicate roughly full capitalization of the tax differentials, but is based on only 86 observations. Palmon

2This is quite different from property taxes being a relatively salient tax to existing house owners, and in relation to other taxes, which is discussed by Cabral and Hoxby (2012).

and Smith (1998) use a similar approach and estimate house price effects from cross- sectional variation in tax rates within a district with similar local public services. Their estimates suggest full capitalization and they conclude that agents on the market rationally incorporate tax liabilities in their valuation of houses. While solving the most critical endogeneity problem, their analysis is also based on a limited data set with only about 500 observations, and the estimates may still contain some bias due to an inability to fully account for differences in unobserved factors, like neighborhood characteristics and amenities, that may be correlated with the tax rate.

Like Cushing (1984) and Palmon and Smith (1998), we solve the simultaneity and measurement problems by estimating capitalization in a context where the property tax rate is unrelated to local public services. The Swedish property tax is determined at the national level and do not vary between municipalities. However, unlike most previous studies, we rely on variation stemming from a national property tax reform affecting the tax payments of house owners differently depending on the assessed tax value of their house. Starting as an election promise in the campaign of the 2006 parliamentary election, the property tax rate was lowered in 2008 from 1 to 0.75 percent and a cap on property taxes was set at SEK 6,000.

As a consequence, a substantial fraction of house owners received a large reduction in yearly tax liabilities. We rely on, and show that our setting is suitable for, difference-in-differences (DID) estimation as the policy change affected houses with previously similar price developments differently. Our analyses are based on a rich data set covering roughly 100,000 sales of owner occupied single-family houses in Sweden during 2006, 2007 and 2008. The data set contains information about tax liability, as well as the exact date when the sales contract was signed, the final price, characteristics of the property, location and more.

We make an important contribution to the previous literature on tax capitalization in that we estimate responses to both announcements and actual changes in the property tax.

To our knowledge we are the first to estimate responses to an announcement of a property tax change. Due to our large and nationally encompassing data set we are also able to study heterogeneity in price responses in several interesting sub-markets and to relate our findings to the growing literature on the empirical relevance of bounded rationality.

Although we cannot provide direct evidence of bounded rationality, we are able to study price responses in sub-markets, where individuals are likely to be more informed about changes in the tax schedule and where the education level is higher (and hence financial

3SEK 6 ≈ USD 1 and SEK 9 ≈ EUR 1.

4The studies most related to ours, using Swedish data, are Berger et al. (2000), finding that government subsidized interest rates on mortgages in the 80’s and early 90’s were fully capitalized into property prices, and Boije (1997) who finds that the local income tax rate is partially capitalized into property prices.

Other related papers that analyze responses in the housing markets to transfer taxation are Best and Kleven (2013) and Kopczuk and Munroe (2014)).

literacy). Since we find significant capitalization estimates in these markets only, we find it plausible that bounded rationality is a key to why we find negligible capitalization rates overall. In this sense our paper also contributes to the growing literature on how cognitive limitations and information costs influence responses to public policies in general, and to complex tax codes in particular (see e.g.Chetty et al. (2009), Kling et al. (2012), Sahm et al. (2012) and Chetty and Saez (2013)).

2 The decision problem - what to pay for a house?

In this section, we present a framework for interpreting our empirical results. We start out by presenting standard capitalization theory and then incorporate insights from models of inattention and bounded rationality, see (DellaVigna (2009) and Chetty (2009)). The typical view in capitalization models is that property prices reflect the present value of future housing services net of costs (see e.g. Sirmans et al. (2008)). This means that prices are determined entirely by the demand for different properties, implicitly (or sometimes explicitly) assuming that supply is inelastic. A simple equation can thus illustrate how prices are formed:

P = S

r + S

r + ... + S

r − C

r − C

r − ... − C

r , (1)

where P is the price of the house, S

is the value of service i, r represents the relevant discount rate, M denotes the number of different services the property provides.

and

thus represent the net present value of service i and cost j respectively, assuming an infinite horizon. In principle, S can denote any aspect of the property that consumers value, like location, size, attractiveness, access to good schools etc. C denotes costs associated with the property and N the number of relevant costs. C can be e.g. capital costs, heating costs, and of course: tax liability, which we will focus on. In principle, this equation could be estimated using the following empirical model:

P = θ

S

r + θ

S

r + ... + θ

S

r − β

C

r − β

C

r − ... − β

C

r (2)

The standard view in capitalization models is that θ

= 1 ∀i and that β

= 1 ∀j, meaning that all aspects of a property are fully valued at their net present value. However, if we take into account the context and the process by which an individual makes a decision about how much to pay for a house, it is evident that optimization errors might occur.

The prescribed view in Simon (1978) is that rationality should be view in a broader sense than the strict sense of maximization that was prevailing in the economic sciences.

A weaker definition of rationality takes into account the decision environment and the

cognitive limitations of the individual decision maker. This weaker form of rationality is

typically referred to as “bounded rationality”. As is also put forward by Chetty (2009) and DellaVigna (2009), it may be boundedly rational to neglect less salient aspects when an individual is in a complex decision making environment. Chetty (2009) presents a framework for how saliency can be incorporated into a model of commodity demand.

Incorporating these insights into our capitalization framework would mean that for some i, θ

< 1 and for some j, β

< 1. If we rank the aspects and costs according to how salient consumers perceive them, with S

and C

being the most salient quality and cost aspects respectively we get the following predictions:

θ

> θ

> ... > θ

and β

> β

> ... > β

We argue that in the Swedish context, the property tax liability was not among the more salient aspects (neither before nor after the tax reduction) when a prospective house buyer is considering how much to pay for a house. Instead consumers are likely to find the aspects that are typically highlighted in advertisements as more salient. The more salient quality aspects would then be location (possibly including access to good schools), living area and plot size. Less salient aspects would perhaps be the attractiveness of the garden in the winter (assuming that it is marketed during summer), behavior of neighbors and occasional disturbance from heavy traffic on a nearby road. Since costs always put downward pressure on the final market price, real estate agents have incentives to not emphasize the costs of living in the house as much as they would emphasize quality aspects. Additionally, we argue that the saliency of a particular type of cost depends on how high the cost is in relation to other costs of owning a house. For an average house and buyer, interest payments on mortgage loans perhaps constitute the largest cost of owning a house. In Sweden, heating and maintenance costs are also very high for many houses. Water, sewage, and insurance costs are typically smaller and hence, we argue, less salient costs. Property tax liabilities could before 2008 be larger than for instance heating costs for very expensive houses, but were for most houses lower, but still higher than, for instance, insurance costs. This makes us think of the property tax as neither the most nor the least salient cost of owning a house. If some consumers do not perfectly consider all these aspects when deciding how much they are willing to pay for a house then we should expect estimates of θ and β to be considerably lower on less salient aspects.

If the price is determined through an auction process, which is common in the market

for residential housing in many countries, then it is crucial whether fully rational or bound-

edly rational consumers are more likely to put the highest bid. If inattentive consumers

neglect a certain cost aspect, then they would be willing to pay a higher price for a house

than fully rational consumers. In such a case, a reduction in that particular cost would

increase the price fully rational consumers would be willing to pay for the house. However,

inattentive consumers would still be willing to pay more (as long as the neglected cost is not reduced to zero), and hence win the auction and determine the price.

Taken together, we argue that it would be quite expected that property taxes would not be fully capitalized, at least in tax regimes were the annual tax liability does not put property taxes on top of the list of housing costs.

3 The Swedish property taxes and the reform

Between 1985 and 2008, the property tax was uniform across the country and provided rev- enues to the national government. Both industrial and residential properties were subject to the tax, although the focus of our study is on owner occupied residential properties.

In 2006, the tax revenues from this group of properties amounted to SEK 13.1 billion, or about 1 percent of total tax revenues. The guiding principle of property taxation in Sweden is that property owners are liable to pay a percentage on the tax value of the property. From 2001 until 2008 the tax rate was 1 percent of the tax value. The tax value is determined by the tax authority and is approximately 75 percent of the market value of the property (at the time of the assessment). Each property belongs to a“value area”

according to characteristics of the common surroundings, and the average market price in each value area is one of the factors used in order to calculate the tax value for a single property. Properties also receive so called “standard points” from the tax authority which take into account individual property characteristics. Since 2001, tax values are reassessed every third year. Consequently, when market values rise, in between assessments, the tax value as a share of market price decrease. Increases in tax liabilities, due to reassessment of tax values, were phased in over a period of three years, such that owners only had to pay one third of the increase in tax liability the first year after a reassessment.

When determining the appropriate property tax rate, an important principle is that it should be neutral in relation to the tax on capital (or rental) incomes. If home owners receive a rental income amounting to 3 percent of the market value of the residential property, a tax rate of 1 percent of the tax value, which is 75 percent of the market value, is equal to a 25 percent capital income tax. This tax rate can be compared with the 30 percent tax on capital incomes in Sweden. In addition, when a property is sold above its purchase price, the seller have to pay a capital gains tax amounting to 20 percent of the profit.

5A “limitation rule” was introduced in 2001, such that households with a yearly income of less than SEK 600,000 would not pay more than 5 percent of their income in property taxes.

Figure 1: The final tax reform 2008 (values in SEK)

- 6







 

800,000

6,000 ^Post

Pre

Tax value Tax liability

(a) Yearly tax liability pre- and post-reform

- 6

800,000 2,000

Tax value Tax Reduc.

(b) Reduction in yearly tax liability

3.1 The 2008 reform

On January 1, 2008, the property tax on residential housing in Sweden was reconstructed.

The tax rate was lowered from 1 percent to 0.75 percent of the tax value and a cap on the yearly tax liability was set at SEK 6,000.

The tax revenues were shifted from the national government to the municipal governments. However, grants from the national government to the municipalities were lowered to cancel out the revenue increase. The property tax reform was thus intended to be neutral in terms of the municipal budget.

Unlike what is common in other countries, Swedish municipalities are not allowed to set their preferred local property tax rate. The tax rate was kept uniform across the country even after the 2008 reform. The average yearly reduction in tax liabilities from the 2008 reform amounted to SEK 4,900, and the average net present value of the tax reduction was SEK 245,000.

With the new property tax system, tax revenues from owner occupied residential properties decreased from SEK 13.1 billion in 2006 to SEK 10.5 billion in 2008.

The tax decrease was partly financed with an increase in the capital gains tax from 20 percent to 22 percent, and partly by the introduction of an interest rate on delays on these tax payments.

In figure 1(a) we show how yearly property tax payments depend on the tax value, both before and after the reform. The straight line with a gradient of 0.01 is the pre-reform tax schedule, and it shows how the tax used to be strictly proportional to the tax value.

The post-reform tax schedule, however, has a gradient of 0.0075 and a kink at SEK 6,000 in yearly tax liabilities, which illustrates clearly that the reform came in two parts; the proportional tax decrease of 0.25 percentage points and the cap at SEK 6,000. Combining

6The limitation rule was kept but would only apply to senior citizens.

7Details about how the net present value is calculated is presented in section 5.2

these two schedules, we get a kink in the yearly gain from the tax reform, which is shown in figure 1(b). The figures illustrates that the tax reduction was larger for highly valued houses. The relationship is piecewise linear in the tax value; the higher the tax value the higher the gain, although all house owners clearly benefited from the reform in some way.

3.2 From promise to implementation

Like in the U.S. and elsewhere, the property tax is unpopular and regularly discussed also in Sweden.

House owners lobby for a complete removal, economists support the tax for efficiency reasons, and politicians want the revenue it brings but also the support of the voters. Between 1994 and 2006, the Social Democrats ruled as a minority government in Sweden. During this time, one of the small opposition parties, the Christian Democrats

repeatedly propagated for a removal or large reduction of the property tax. The other parties in the opposition were also somewhat positive towards a reduction of the property tax, but it took until 2006 before the opposition jointly decided to give it priority.

In the summer of 2006, a few months before the election in September, the newly formed coalition of center-right opposition parties – the Alliance for Sweden (Swedish:

Allians f¨ or Sverige) - jointly announced a large reduction of the property tax as one of their most tangible election promises. On 4 July, The Alliance parties summoned a joint press conference at the Almedalen Week (a highly publicized annual conference where interest groups, media representatives and political parties discuss politics) to announce their agreement on the future of the property tax. The proposal was in two parts: in 2007, the tax values were to be fixed at the 2006 level and the yearly tax liability on the land part of the property should not exceed SEK 5,000. In 2008, the national property tax was to be replaced with a “low” property tax collected by the municipalities. The Alliance parties did not say how high the tax would be, but the Christian Democrats’ proposal at the time was a cap at SEK 2,800 in yearly tax payments. No change in the tax rate was mentioned. It was also clearly communicated that the property tax should not increase for anyone, and that it should be partly financed within the housing sector.

After the announcement, economists appeared in the media, defending the tax. Social Democratic party representatives deemed the proposal as irresponsible due to the lack of a financing plan of the tax cut, and also pointed out that the proposed tax reduction would predominantly benefit owners of highly valued properties. The Social Democratic response to the Alliance proposal was that the system should stay mainly as it was, although a reduction of the tax rate was on the table.

In the election manifesto, released on 23 August, 2006, the promise to abolish the

8See Cabral and Hoxby (2012) for a discussion on the unpopularity of the U.S property tax.

9The Christian Democrats received roughly 9 and 6 percent of the votes in the 2002 and 2006 parlia- mentary elections.

property tax in its current form was repeated. In the manifesto, The Alliance referred to the tax as being “unjust” and “unpredictable”, and they expressed an ambition to abolish the tax by 2008. Until then, they wished to implement the cap on the land part of the tax at SEK 5,000 and keep tax values at the 2006 level, as earlier announced at the press conference in July. On 17 September, 2006, the Alliance won the election and formed a majority government. In connection with the inauguration, the new government repeated the long-term promise to reform the property tax, and also the short-term promise to reduce the tax liability through the implementation of the temporary land tax reform.

The new government implemented the promised temporary land tax reform on 1 Jan- uary 2007.

The land tax reform was estimated to cost SEK 2.9 billion in lost revenues.

During 2007 there was a public discussion about where the final cap should be set. On 20 September the final proposal of a cap at SEK 6,000 and a tax decrease to 0.75 percent was finally publicly announced. On 25 October the proposition was put forward, which was later on accepted by the Parliament on 17 December. Figure 2 shows the timing of the reform. Since 2008, the cap has been raised according to increases in general income.

Figure 2: Timing of the property tax reform

-

1 Jan 2006 1 Jan 2007 1 Jan 2008

?

4 Jul

First announcement of the property tax reform.

?

17 Sept

The Alliance wins the election.

?

1 Jan

Temporary land tax reform is implemented, where the owner pays either SEK 5000, SEK 2/M², or 1 percent of the tax value of the land part of the property.

?

25 Oct

Government presents bill with final reform, which passes on 17 Dec in the parliament.

?

1 Jan

The new property tax rate is implemented, which is 0.75 percent of the tax value but with a cap at SEK 6,000

10The land tax reform affected taxes retrospectively for 2006 as well, but it was not part of the promise and could not easily have been anticipated by the agents in the housing market. Thus it should not affect our capitalization estimates.

3.3 What could prospective house buyers and sellers expect?

On 4 July 2006, when the Alliance jointly announced that they would substantially reduce the property tax, it was a clear break from what voters could have expected earlier about changes in the property tax. Before that date, a major reduction in the property tax was only given priority by the Christian Democrats – a small opposition party. However, it was expected that the election would be a close race and even if the Alliance would win, it would still not be certain that the promise would be delivered on. In a highly efficient market with rational agents, this kind of news would still immediately increase asset prices, at least to some degree. After the Alliance won the election and re-announced that they would implement their promise to reduce the property tax, the probability of a reduction again increased. Finally, from 1 January 2008, when the new tax schedule was implemented, the only uncertainty that remained was whether the Social Democrats would revert the property tax, were they to win the election in 2010. However, in the summer of 2009 it became evident that the Social Democrats would only like to increase the property tax for houses with very high tax values.

House owners and prospective buyers, therefore, had good reason to believe that the property tax reduction would not be reverted for a long time. Still to date, only the Left Party, which is in the opposition and has less than 6 percent of the votes, has proposed to make substantial increases in the property tax.

4 Data

We have at our disposal a combined data set originating from the Swedish land survey- ing office and Svensk M¨ aklarstatistik AB (Our translation: Swedish Real Estate Agent Statistics, Inc.). The data set contains 124,563 observations of instances where a house – intended for permanent or summer living by the owner – has switched owners in the land register during the time period 2006 to 2008, for all of Sweden. We have the following information about the properties included in our data set: date of contract signing, mar- ket price, tax value, zip code, year of construction, and other housing characteristics such as living and plot area. The data set covers 41 percent of the total population of owner changes in the Swedish housing market. The reason why we do not cover the whole pop- ulation is that we lack information on sales of properties mediated without a real estate agent and agents that are not connected to Svensk M¨ aklarstatistik AB (mainly smaller real estate agents), and thus we cannot include these sales in our sample.

In the empirical analysis we will mainly work with a sample of 101,449 observations.

11In a leading Swedish newspaper the Social Democrats wrote that they would like to implement an additional property tax amounting to 1 percent on house values above SEK 4.5 million. Dagens Nyheter, 30 June 2009.

We leave out 23,114 observations due to the following deliberate sample restrictions. First of all, we leave out a few (131 observations) owner changes that are due to inheritance, distribution of marital properties, premarital settlements, gifts, purchases between family members and the like. Second, we only keep observations that refer to one singular tax value unit, and which contains exactly one house and exactly one land unit (2,476 obser- vations). Third, only properties that had their latest tax value assessed in the mandatory tax value assessment in 2006 are included in the sample since tax liability values as a rule depend on market prices in the value area and tax value assessments after 2006 will therefore be endogenous to the tax reform (1,005 observations). Assessments in 2007 and 2008 were only conducted for rebuilt houses. Fourth, we also drop observations where:

the purchase concerns leases instead of actual ownership (5,727 observations), and cases where the tax value is not completely related to the property that has been sold (1,551 observations). A few observations lack information on value area, value year, tax value, living area and plot area, and these observations are therefore dropped (221 observations).

The most extensive sample restriction (12,003 observations) is due to the exclusion of the second half of 2008, when very few houses were sold especially at the very end of the year, due to the extraordinary event of the financial crisis.

4.1 Descriptives

The distribution of tax values in our data set is shown in figure 3. The distribution is positively skewed, with a long tail of relatively highly valued properties. A majority of properties have a tax value below SEK 800,000 (52 percent), and were affected by the reform only to the extent that the tax rate was decreased from 1 percent to 0.75 percent of the tax value. Owners of properties with a tax value above SEK 800,000 received a more favorable treatment from the tax reform because the marginal tax rate above SEK 800,000 was set to zero.

In table 1 we show some descriptives of the main variables used in the data analysis.

The average house in the data set costs SEK 1,778,000, and has a tax value of SEK 959,000.

The average net present value gain from the final 2008 reform is SEK 245,000. In terms of yearly tax payments, the average gain is SEK 4,900. The average house has 115 m

of living area, and has a plot size of 1500 m

. The so called “standard points” is a point system used by the tax authority in order to estimate the tax value taking into account several additional house quality characteristics such as: tiled stoves, kitchen appliances, and the number of bathrooms. The standard points of the properties in our sample range from 6 to 54 with a mean of 29.

Figure 4a shows the evolution of markets prices and figure 4b the number of sales over

the time period that we study. They show both the actual time series and seasonally

Figure 3: Distribution of tax values

0200040006000Frequency

1 500 1000 1500 2000 2500 3000

Tax value

Tax values are in SEK 1,000

Table 1: Descriptive statistics

Mean Std.dev. Min Max

Market price (SEK) 1,778,000 1,330,000 10,000 24,000,000

Tax value (SEK) 959,000 692,000 55,000 11,576,000

Net present value gain (SEK) 245,000 295,000 7,000 5,488,000

Yearly gain (SEK) 4,900 5,900 138 110,000

Living area m² 115 41 12 1,680

Total area m² 1,500 2,500 25 162,000

Standard points 29 6 6 54

Monetary values are in nominal prices. All values are rounded. 101,449 observations.

adjusted series. The fluctuations in the sales prices indicate a monthly seasonal pattern.

House prices fall during midsummer and winter, while they increase during spring and

autumn. This seasonal pattern is due to the type of houses sold, and the number of sales

in each month. In both cases it is clear that market prices show an upward trend, which

is broken by the end of 2007. There is also a clear seasonal pattern in figure 4b which

shows the number of sales each month. The sales quantity is pretty stable over time until

the financial crisis in 2008 when the number of sales decreases sharply.

Figure 4: Time series

100015002000Thousand SEK

2006m1 2006m7 2007m1 2007m7 2008m1 2008m7 Actual Seasonally adjusted

(a) Market price

100030005000Number of sales

2006m1 2006m7 2007m1 2007m7 2008m1 2008m7 Actual Seasonally adjusted

(b) Number of sales

5 Empirical strategy

As was shown in figure 1, yearly tax liabilities are a deterministic function of the tax value, both before and after the reform. The tax value is of course correlated with the market price through the characteristics of the house. We also know that the reduction in tax liability is larger for more expensive houses. A simple cross-sectional regression of market price on tax reduction would therefore yield a positive coefficient as a consequence of the construction of the tax reform itself. In other words, (tax reduction) treatment is correlated with house characteristics and is therefore endogenous. Since we have data on house sales both before and after the reform, we can deal with this endogeneity issue in a Differences-in-Differences (DID) framework.

We use three different DID approaches, where the first one is a conventional two-group DID estimation where we compare the evolution of market prices of properties for which the cap is binding, with properties that were only subject to a decrease in the tax rate.

In other words, the price development of properties with tax values below SEK 800,000

(control group) serves as counterfactual for the price development of properties with tax

values above SEK 800,000. In the second approach we divide the treatment group into

three parts according to tax value. By doing this we allow for heterogeneity in treatment

effects in line with predictions from capitalization theory and from how the reform is

designed.. We can also observe whether properties with higher tax values are closer to

their theoretical effects than properties with lower values, which is what we would expect

if saliency of the property tax increases in the magnitude of the tax liability. In the final

approach, we utilize the full extent of variation that we get from the reform and perform a

DID with continuous treatment. With the latter approach we can estimate the relationship

between the tax reduction and the price response, taking into account that each property

received a unique treatment.

5.1 Specification of empirical models

In the two-group DID case we estimate variations of the following empirical model

y

= P

+ T

+ D

+ ΩX

+ V

+ ε

, (3) where y

is the natural logarithm of the market price for house i, in group g (g = 1 for the treated group and g = 0 for the control group), region j, and in period t. We include period specific dummies in P

and group specific dummies in T

(T

equals 1 if g=1 and 0 otherwise). The parameter(s) of interest are D

, which are interaction variables between the group dummy (T

) and period dummies (P

), D

= (T

×P

). The interaction variables D

equal 1 if the house is in the treated group in period t, and is 0 in all other cases. D

represents the effect of the tax reform in period t = k. X represents a vector of control variables. We also estimate the model with fixed effects V

at the county or municipality level.

The important assumption underlying all DID procedures is the parallel trend as- sumption. Given that we have chosen to specify our model in logarithms, we assume that properties in both the control and treatment group would evolve in the same way in percentages in the case with no property tax reform. If properties on all price levels are capital investments yielding the same rate of return, this is more plausible than assuming parallel trends in the market price level. This assumption is not directly testable. How- ever, we show, in Figure 6 in Section 5.3, that prices of houses above and below the cap evolve similarly before the first announcement in Almedalen.

To somewhat relax the parallel trend assumption and to increase the precision of the estimated treatment effect we include control variables for the size of the house (living area), the size of the land property (plot area), and a measure of the quality or standard of the house (standard points) in X

. When including control variables, we rely on the slightly weaker assumption that the two groups would have evolved equally conditional on the control variables. Including fixed effects in the models further allows us to deal with unobserved heterogeneity, and should thus decrease potential bias in the estimated price responses. However, once we include fixed effects, we utilize variation only within specific areas to identify price responses. This comes with the cost that several fixed effects unit will not contribute with identifying variation because they contain properties from the treatment only or the control group only. In practice this means that the theoretically

12Given the kink in the treatment at SEK 800,000 in tax value, it would be natural to use a regression kink design to identify the causal effect of the tax reduction. We have tried such a strategy but obtained estimates that were to imprecise to be informative. This is not surprising given that the difference in the tax reduction, and hence the expected difference in price response, is quite small around the kink.

expected estimates under full capitalization will be lower. This is because the variation in tax values, in areas with properties from both groups, on average is smaller than the variation in the country as a whole. We acknowledge this when we calculate capitalization degrees and interpret our findings.

In the two-group DID case we assign properties below the cap (0-800’ in tax values) to the control group, and properties above the cap (>800’) to the treatment group. In the four-group DID case we split the treatment group into three groups according to the following tax values, 800’-1600’, 1600’-2400’ and >2400’. The empirical model looks the same as in equation (3), although we now have three dummy variables in T

.

In the case with continuous treatment we estimate the following empirical model

y

= P

+ T

+ τ

+ (P

× T

) + (P

× τ

) + (T

× τ

) + V

+ D

+ ΩX

+ ε

, (4) Where D

= (P

× τ

× T

), and τ

is the natural logarithm of the tax value. In other words, the model is similar to a DDD (triple differences framework). We again rely on a logarithmic specification, in both market price and tax value. The expected price response under full capitalization is concave in the tax value for houses with tax values above 800’ , i.e. houses with higher tax values should get a higher percentage increase in house prices from the reform, but this effect is diminishing. The concave relationship is illustrated in figure 5a, where it is also clear that the properties below the cap all got the same proportional treatment.

Figure 5: Continuous treatment

0.1.2.3NPV/Market price

0 1000 2000 3000

Tax value

(a) Tax value

0.1.2.3NPV/Market price

4 5 6 7 8 9

Ln Tax value

(b) Logarithm of tax value

Figure 5a illustrates that the relationship between market price and the theoretically

expected price response (see next subsection) under full capitalization. As can be seen in

Figure 5b, the logarithm of the tax value and the price response in the natural logarithm

of the price is approximately piecewise linear. Using a log-log specification we can then

expect the effect of the property tax to be linear in the tax value for houses that are

above the cap. For houses below the cap there is no variation in proportional treatment and hence expected proportional price responses. The houses below the kink are not used to identify the effect of the property tax when using continuous treatment, but can be included to increase the precision of the control variables. The parallel trend assumption under the case with continuous treatment is that the relationship between log market price and log tax value would be the same for all tax values without treatment. In order words, property prices evolve in the same way – in percent – for all levels of the log tax value.

We split our sample period of two and a half years into five time periods P

, each of six months:

• Control period (Jan 2006 - Jun 2006): No annoncements had been made

• First announcement period (Jul 2006 - Dec 2006): Announcement at Almedalen and election win by the Alliance.

• Land tax reform period (Jan 2007 - Jun 2007): The land tax reform is effective

• Final announcement period (July 2007 - Dec 2007): The final reform is announced and passed by the Parliament

• Post reform period (Jan 2008 - Jun 2008): The final reform has been implemented The above split allows us to study price responses in all important stages of the reform, from the first election announcement in Almedalen in 2006, to the final implementation of the reform in 2008. If house buyers anticipated the final reform already before the implementation, we would expect to see a jump in the house price series of highly valued houses already before the implementation of the final reform. Such responses could occur after the first announcement in Almedalen, after the Alliance won the election, or when the final reform was announced in the second half of 2007. If the final implementation came as a surprise we should see responses only in the last period.

5.2 Expected price responses under full capitalization

If the tax decrease is fully capitalized – in what order of magnitude do we expect the

estimates to be? In this section we calculate theoretical estimates that are expected when

the full value of the tax decrease is capitalized. We begin by calculating the net present

value (NPV) from the 2008 reform for each property (as detailed below) based on its

tax value. We add the NPV to the actual market price in order to get a market price

under full capitalization. We then perform a simulated DID estimation using both the log

market price under full capitalization as outcome variable and the actual market price. In

the case with two-group DID, we first take the difference between log market price with

treatment and the actual market price within each group separately. Lastly we take the second difference, the difference-in-differences, between the two groups in order to arrive at an estimate which theoretically would emerge under full capitalization. This procedure perfectly replicates the empirical specification in the previous section, which is important in order to get a credible comparison between our empirical and theoretical estimates.

Note that when performing these calculations, we restrict the sample to the pre-treatment period of the first six months of 2006, as the market prices of these properties are not affected by the reform. Also, in the empirical analysis the difference in the first half of 2006 constitutes our baseline comparison.

Table 2: Theoretical effects under full capitalization

Reform effect Theoretical DID-estimate

≤800’ 0.074 control group

>800’ 0.155 0.081

800’-1600’ 0.130 0.056

1600’-2400’ 0.194 0.120

>2400 0.224 0.150

The theoretical DID-estimates under full capitalization are presented in table 2. The first column displays the reform effect for each group separately, expressed in log differ- ences. Under full capitalization, the proportional tax decrease for the control group, the

≤800’ group, implies an increase in prices of approximately 7.4 percent. For properties above the cap, the >800’ group, the reform implies a 15.5 percent increase in prices.

Dividing the treated group into three new groups, we can clearly see that the reform ef- fect is increasing in the tax value. The second column displays the second difference, in other words, the theoretical DID estimates for the various groups. In the two-group DID- estimation we expect to see an estimate of 0.081 under full capitalization. The treated group is expected to increase by 8.1 percentage points more than the control group due to the reform. In the four-group DID we expect to see the estimates 0.056, 0.120, and 0.150 respectively.

In the case with continuous treatment we first estimate the pre-reform relationship

between the log tax value and the log market price. We then add the NPV gain of the tax

reduction to each property and re-estimate the relationship. The difference in the slope

coefficients between the two models reveal the expected continuous DID estimates under

full capitalization. We stated earlier that there was no variation in proportional treatment

among the properties below the cap. This statement is confirmed in table 3 where we show

that adding the NPV of the reform does not change the relationship between log market

price and log tax value in this group of properties. The reason is of course that the

treatment from the reform for these properties was a proportional tax decrease. For the

properties above the cap we expect to see a change in the relationship between the log tax value and log market price according to table 3. For the group of properties with tax values above 800’ the coefficient on log tax value from a regression on log market prince will increase from 0.94 to 1.06, which implies an expected theoretical continuous DID-estimate at 0.12.

Table 3: Theoretical continuous DID-estimates under full capitalization

Before treatment After treatment Theoretical DID-estimate

≤800’ 1.00 1.00 0.00

>800’ 0.94 1.06 0.12

Note: Column 1 and 2 show the relationship between log tax value and log market price.

The theoretical estimates depend on assumptions about the agents’ expected develop- ment of the property tax, time horizons and discount rates. A short reflection on each of these assumptions is needed. First, we note that the net present value of a stream of incomes can be written as:

N P V =

T

X

i=0

I

(1 + r + π)

, (5)

were I

denote the income in period i, or in our case the annual tax reduction, r the real interest rate and π the inflation rate. The tricky part is to calculate I

. According to the details in the property tax bill effective from 1 Jan 2008, the cap which was initially set at 800’ SEK would be adjusted on a yearly basis according to both inflation and the real growth in wage incomes (g). Assuming that tax values (and hence tax payments under the old property tax regime) also increase at the rate π+g, we get I

= I

× (1 + g + π).

With the simplifying assumption that the real interest rate equals the growth in real wages we get that:

N P V = I

× T (6)

The value of the tax reduction thus depends critically on the time horizon T . With no future changes in the tax policy, the time horizon should be very long. In standard net present value calculations, the flow of incomes is certain. However in our case, the property tax is likely to change again at some point in time. Both further reductions and increases are possible. This means that a rational agent may discount the gain from the reform with a risk premium. We assume a risk premium of 2 percent and an infinite discount horizon. This is equivalent to a case with no uncertainty and that the property tax is reverted after 50 years.

Our NPV calculations assume that the agents understand the details about how the

tax payments would evolve, both under the old and new policy, and that they adjust

for the political uncertainty. If we instead assume that agents simplify the problem and expect the first year’s reduction to be constant in real terms, we would obtain exactly the same NPV, with a real interest rate of 2 percent.

The most critical assumption underlying the theoretical DID estimates is the choice of time horizon (or political uncertainty). But the choice of real interest rate is also important. The theoretical estimates would be smaller with a shorter time horizon and with a higher real interest rate.

5.3 Validation of identifying assumptions

As we have shown in figure 1, the treatment from the reform increases in the tax value. In the following empirical analysis we will compare the price developments of properties that were subject to both the tax rate reduction and the cap with properties that were subject to only the tax rate reduction. Since the control group also gained from the property tax reform, the counterfactual will be the case of a proportional decrease in the property tax.

Descriptives for the two groups are shown in in table 4. Naturally, the average market price and the average tax value are higher in the treated group. A property in the control group has an average tax value of about SEK 462,000 and was on average sold for SEK 903,000, whereas an average property in the treated group has a tax value of SEK 1,497,000 and was sold for SEK 2,726,000. The houses in the treated group are 31 m

larger than in the control group, and they are also of higher standard, as reflected in having more standard points. The average plot size in the control group is 1800 m

, but only about 1100 m

in the control group. This difference follows from the fact that more properties in the control group are located in the countryside where plot sizes are larger than in urban areas. The group sizes are well balanced.

Table 4: House characteristics by treatment status

Control Treatment

≤800 >800

Mean Mean

Market price (SEK) 903,000 2,726,000

Tax value (SEK) 462,000 1,497,000

Yearly gain (SEK) 1,200 9,000

Net Present Value (SEK) 58,000 449,000

Living area m² 100 131

Total area m² 1800 1100

Standard points 27 31

Observations 52,750 48,699

SEK in nominal prices. All values are rounded.

A DID approach allows for differences in the characteristics of the treated and un- treated group, and therefore also in the levels of the outcome variable. Instead, a DID strategy relies on the assumption of parallel trends, i.e. that the market price in the treated group would evolve according to the control group in the case of no treatment. In figure 6 we show (monthly) seasonally adjusted market price indices by group status. The solid line represents the control group and the dashed line represents the treated group.

The two groups evolve according to a seemingly similar trend until mid 2007, when the market price in the control group starts increasing slightly faster than in the treated group.

Already with this simple descriptive illustration we can see patterns that are not expected under the prediction from capitalization theory. Prices of expensive houses do not seem to increase more than lower valued houses, neither before nor after the reform. Under full capitalization, and no pre-reform responses, we would expect to see the prices of the treatment group to increase with 8 points relative to the control group by 1 January 2008.

This is clearly not visible in figure 6.

Figure 6: Seasonally adjusted indices by group

8090100110120130

2006m1 2006m7 2007m1 2007m7 2008m1 2008m7

0−800’ >800

Seasonal adjustment: log market prices were regressed on a full set of calendar monthly dummies, for each group separately. The mean of the series was added to the residuals from this regression.

6 Results

We now turn to the empirical estimates of the price responses of the property tax reduction.

We start by presenting our baseline results from the two-group DID model. These results

show the average treatment effects of the average treatment difference between the control

and treatment group. Moreover, we also show the dynamics of price responses - i.e. pre-

reform (or announcement effects), as well as post-reform responses. We then continue by studying responses using three stratified treatment groups according to tax value. We thus allow for treatment heterogeneity in line with the heterogeneity in expected theoretical effects, see table 2. Thereafter, we present continuous DID estimates, which allows us to exploit the full extent of variation in treatment. Moreover, we discuss the robustness of our findings with respect to different assumptions about expectations, and the possible influence of other policy changes occurring during our sample period. Finally we present estimates of price responses in several interesting sub-markets.

6.1 Two-group DID

In table 5 we show the results from our baseline DID estimations using two groups and five time periods. The control group consists of properties below the cap, and the treatment group consists of properties above the cap. The columns display the results from run- ning different variations of equation (3). The group dummy and time period interaction variables D

are named after the most important reform event happening within the six month time period. Model 1 is the simplest version of equation (3), where there are no controls or fixed effects, in model 2 we add property characteristics as controls, and in model 3 and 4 we add county and municipality fixed effects respectively.

It is certainly not trivial to know what expectations of the property tax reform buyers might have had in the “First announcement” period. Although, keeping in mind that this period covers both a generally stated election promise and a subsequent win by the Alliance for Sweden, it would not be a stretch to say that we expect to see at least some announcement or anticipation effect at this point. However, the results show that there are no substantial differences in the price changes between the two groups in this period. The statistically insignificant estimates in the four models range between −0.5 and +0.7 p.p difference in price development. These results should be compared with the theoretically expected estimates which are 7.9 p.p for model 1 and 3.9 p.p for model 4 under full capitalization of the 2008 tax reform and perfect foresight.

In the following period, going into 2007, the land tax reform is in place and the details of the final reform is being discussed. Neither in this time period do we see any clear indications of capitalization of the property tax, although estimates are somewhat larger than in the previous period. The highest estimate of 1.5 p.p is found in model 3, an effect size which should be compared with the theoretical estimate of 5.6 p.p. Hence, none of the estimates for the “Land tax reform” period are statistically significant or economically substantial.

The third six-month period is when the theoretical assumptions about the final reform

completely converges with the expectations of the voters. When the reform is announced

in it’s final form, we expect to see clear capitalization effects since there is no longer any uncertainty regarding the details of the reform. What we see in the results however, are even smaller estimates than in the previous period. The additional certainty about the coming reform has accordingly not caused the price dynamics between the two groups to diverge, rather the opposite, they converge to the same proportionality as in the beginning of 2006.

Estimates in the last “Post reform” period, when the reform is a law in effect, are negative and hence in the opposite direction of what we expect from our theoretical as- sumptions. The estimates range from −2.5 p.p to −1.4 p.p, and they should be compared with the theoretical estimates of 3.9 p.p to 7.9 for each model respectively.

Taken together, the results from our baseline DID-estimation using two groups provide no evidence of capitalization of the property tax reduction. From a theoretical perspective we expect all estimates to be positive, but to grow over time as the uncertainty about the finally implemented tax change decreases. Instead, we find very small or no effects during the three periods leading up to the “Post reform” period where we, opposite to what we expected, find negative effects but substantially small effects.

Table 5: Two-group DID

(1) (2) (3) (4)

First announcement 0.00631 -0.00494 0.00732 0.00631

(0.012) (0.011) (0.010) (0.008)

Land tax reform 0.0127 0.00471 0.0147 0.0123

(0.012) (0.011) (0.011) (0.009)

Final announcement -0.00349 -0.00883 0.00580 0.00546

(0.011) (0.011) (0.010) (0.009)

Post reform -0.0209 -0.0249^∗ -0.0144 -0.0151

(0.012) (0.012) (0.011) (0.009)

Theoretical estimate 0.081 0.073 0.057 0.039

Observations 101449 101449 101449 101449

Adjusted R2 0.584 0.613 0.530 0.427

Controls X X X

Municipality fixed effects X

County fixed effects X

Dep. var: Log House Price. Clustered standard errors on municipality level in parentheses. *p<0.05, **p<0.01,

***p<0.001. 52,750 observations in the control group, and 48,699 observations in the treatment group.

The observant reader will notice that we have a small problem of decreasing power as we add controls and municipality fixed effects. The worst case is found in model 4 in the

“Land tax reform” period, where we cannot reject the null hypothesis of no capitalization

and not reject an estimate of 3 p.p, which could – given some weaker assumptions about

the information set voters has at time – be a reasonable estimate to expect. However, we

can well and good reject full capitalization given the theoretically expected estimate of

3.9 p.p. Of course, we make our conclusions based on the whole set of estimates and over all time periods, and taken together they point in the direction of no effects, at least this far in our empirical analysis. However, we still attempt to solve our power problem by introducing treatment heterogeneity such that the treatment dose is allowed to increase for some groups of properties.

6.1.1 Four-group DID

In the four-group DID model we divide the treated group into three groups according to

intervals of 800’, and perform the same estimations as in the case with two groups. The

estimates in table 6 show the price development in each treatment group (and in the said

time period) relative to the control group. Since the tax reform is designed in such a way

that highly valued properties receive a larger treatment, we are in effect increasing the

treatment dose for these properties. If saliency is important for explaining why we find no

capitalization results in our baseline estimations, we expect to see that estimates for the

highly valued properties are closer to their theoretical estimates than cheaper properties.

Table 6: Four-group DID

(1) (2) (3) (4)

First announcement × (800’-1600’) 0.0186 0.0105 0.0132 0.00942 (0.010) (0.011) (0.010) (0.009) Land tax reform × (800’-1600’) 0.00573 0.00145 0.00822 0.00584 (0.010) (0.010) (0.010) (0.009) Final announcement × (800’-1600’) 0.0134 0.00727 0.0120 0.00696 (0.010) (0.010) (0.010) (0.009) Post reform × (800’-1600’) -0.0108 -0.0137 -0.00810 -0.0141 (0.011) (0.011) (0.011) (0.009) First announcement × (1600’-2400’) 0.0246

0.0138 0.0158 0.00922 (0.011) (0.012) (0.012) (0.011) Land tax reform × (1600’-2400’) 0.0224 0.0151 0.0152 0.0106

(0.012) (0.013) (0.013) (0.013) Final announcement × (1600’-2400’) 0.0204 0.0135 0.0115 -0.000987

(0.012) (0.013) (0.013) (0.011) Post reform × (1600’-2400’) -0.00757 -0.0137 -0.0134 -0.0179 (0.013) (0.013) (0.013) (0.011) First announcement × (>2400’) 0.0404

0.0248 0.0309 0.0148

(0.017) (0.018) (0.017) (0.015) Land tax reform × (>2400’) 0.0709

0.0625

0.0720

0.0660

(0.019) (0.020) (0.020) (0.019) Final announcement × (>2400’) 0.0800

0.0792

0.0828

0.0770

(0.020) (0.020) (0.020) (0.019)

Post reform × (>2400’) 0.0422 0.0383 0.0429 0.0305

(0.025) (0.026) (0.025) (0.020) Theoretical estimate 800’-1600’ 0.057 0.054 0.048 0.040 Theoretical estimate 1600’-2400’ 0.120 0.116 0.104 0.092 Theoretical estimate >2400’ 0.150 0.143 0.127 0.115

Observations 101449 101449 101449 101449

Adjusted R2 0.661 0.673 0.578 0.455

Controls X X X

Municipality fixed effects X

County fixed effects X

Dep. var: Log House Price. Clustered standard errors on municipality level in parentheses. *p<0.05, **p<0.01,

***p<0.001. Group sizes from control to highest valued treatment groups: 52,750; 32,584; 12,092, and 4,023.

What we observe in table 6 is somewhat in line with the reasoning above. Estimates are

higher for properties that receive larger treatment from the reform. Although, this is only a

stable result concerning a small group (4,023 properties) of top valued properties with tax

values above SEK 2,400,000. Already in the first announcement period do we find positive estimates in the range 1.5 p.p to 4 p.p. Estimates increase in the following periods, where the null hypothesis of no capitalization is rejected, and it becomes increasingly difficult to reject full capitalization. In the final announcement period we find estimates in the range of 7.7 p.p to 8.3 p.p, and we are not able to reject full capitalization in any of the four models. In the last period we find positive effects, but they are not statistically significant, and neither are they in the same effect size as in earlier periods. Additionally, we also see that the estimates for the top valued properties are closer to what we would theoretically expect, which is a result that is in line with the saliency explanation.

6.2 Continuous DID

In this section we make use of the whole range of variation and estimate a continuous relationship between reform treatment and market price. Instead of interacting the period dummies with group dummies, we now interact with the (logarithm of) the tax value. In other words, the proportional relationship between tax value and market price is allowed to change over time. In the case of no reform we rely on a continuous version of the parallel trend assumption, namely that proportional trends should be the same for all levels of the log market price. The proportional relationship is theoretically expected to change due to the reform according to what we have shown in figure 5b. Since only properties above the cap have variation in expected treatment effects, we fully interact with a dummy indicating whether the property is below or above the cap.

The results from four models with varying specifications and samples are shown in table 7. The first column with model 1 displays results from estimations using the whole sample and the DDD approach described above. In the first period we get a negative estimate of −1.5 p.p, but we find higher and relatively stable estimates around and above 3 p.p in following time periods. These finding should be compared with the theoretically expected change in the proportional relationship between treatment and market price which is 0.12.

The results remain largely stable when we add controls and municipality fixed effects in

model 2. In model 3 we perform a robustness check by excluding properties with tax values

below 800’ and we estimate a regular DID with continuous treatment. If our assumptions

hold, and there is no proportional treatment effect for properties below 800’, we should

not observe very different results comparing with model 2. Estimates remain relatively

intact in model 3, although the standard errors decrease somewhat. Remembering earlier

analyses where we found highly values houses to be closer to their theoretical effects, we

also perform an analysis when excluding the top valued properties. According to the

results in model 4, the positive results in the three first columns with treatment effect

around 3 p.p seemed to be largely driven by the top valued properties, that received a